Mapping the Integration of AI Literacy Into Information Literacy Instruction in Academic Libraries: A Systematic Review of Emerging Research

Synthesis: Mirza and Arif ask how AI literacy has been folded into information literacy instruction in academic libraries, and find a field still at what they call an infancy stage. Their PRISMA review screened 2,126 records across seven databases and kept 32 studies, 30 of which appeared in 2024 or 2025. Integration is lopsided: 11 studies come from the United States, 19 are quantitative surveys with small or self-selected samples, and the actual teaching runs through LibGuides, tutorials, one-off workshops and revised course modules rather than a curriculum. The authors record alignment with information-literacy frameworks as inconsistent and describe their conclusions as provisional patterns rather than settled findings.

Key Findings

1. The search returned 2,126 records from seven databases; 835 duplicates were removed, 1,140 records were excluded at screening, 151 went to eligibility assessment and 32 articles were included.
2. The included literature is recent and shallow in time: 19 studies were published in 2024, 11 in 2025 and 2 in 2021; none appeared in 2020, 2022 or 2023.
3. The United States contributes 11 of the 32 studies, all published in 2024–2025; the remaining 21 are spread over 14 countries, nine of which contribute one study each.
4. Methods are dominated by quantitative surveys (9, 28%), followed by conceptual papers (7, 22%), mixed methods (6, 19%), qualitative work (5, 16%), viewpoint or descriptive pieces (3, 9%) and reviews (2, 6%).
5. Single-author studies are the largest authorship category (10), yet 17 of the 22 multi-authored studies (about 77%) are inter-institutional.
6. Two journals carry roughly a third of the evidence base: the Journal of Academic Librarianship (6) and IFLA Journal (5).
7. For a large block of studies the review table records no explicit integration with an information-literacy framework; where alignment exists it runs through ACRL, EDUCAUSE, IFLA, DigComp, Chinese IL standards or an ACRL–ACM ethics crosswalk.
8. Challenges cluster into four families: technological and resource constraints (licensed tool access, outdated infrastructure, funding), institutional gaps (no strategy, weak leadership, no policy), pedagogical and human capacity (staff confidence, no structured professional development, no assessment tools) and ethical concerns (algorithmic bias, privacy, academic integrity, overreliance).
9. Emerging practice is workshop- and LibGuide-shaped: integration runs through LibGuides, tutorials, workshops and revised course modules, with named design models including B-Wheel, a Mandala Chart, a research-validation template for cross-examining chatbot output, and a four-level model of awareness, application, evaluation and ethical advocacy.
10. Survey samples are small and self-reported: 760 US library employees, 914, 816 and 632 LIS students in South Asia, 124 and 120 Pakistani librarians, 118 MLS students and 82 Zambian professionals, several with acknowledged self-report bias.

Why "integration" is the wrong verb yet

The review's most useful contribution is negative. It shows that AI literacy has not been integrated into information-literacy instruction in the sense that a curriculum document would recognize; it has been appended to existing delivery formats. A LibGuide page, a workshop session and a revised module are all containers that predate AI, and putting AI content in them leaves the sequence, the learning outcomes and the assessment untouched.

That distinction matters for how the evidence should be read. A study reporting that librarians added an AI segment to an existing course has measured an addition, not a redesign. Where the review finds named models — a validation template, a four-level progression — those are closer to design work, but they are proposals rather than evaluated interventions. The authors' own confidence language tracks this: two research questions are described as strongly supported, while the rest rest on emerging and context-specific evidence.

The geography of a young field

Eleven US studies and nine countries contributing a single study each is a description of where funding, publishing capacity and professional infrastructure currently sit. It is also a caution against generalizing the barriers. The four challenge families are not equally distributed: infrastructure and licensed-tool access dominate in settings where both are scarce, while policy absence and staff confidence dominate where tools exist. A review that aggregates them cannot say which constraint binds in a given library, and its recommendations — structured professional development, institutional strategy, assessment tools — are correspondingly general.

What this means for practice

• Treat a LibGuide as a start, not an outcome. Adding an AI page changes what is available, not what students can do; the review's evidence is consistent with additions that were never assessed.
• Write the learning outcome before the workshop. The weakest competency in the parallel survey literature is designing objectives and assessing AI literacy; a session without an outcome inherits that gap.
• Name a framework and map the content to it. Most included studies record no explicit framework alignment; choosing ACRL, DigComp or a local standard gives the instruction something to be accountable to.
• Check which barrier actually binds. Infrastructure, policy and staff confidence are different problems; a plan that addresses all three at once usually addresses none.

Limitations

• The review is confined to 32 peer-reviewed English-language articles from seven databases covering 2020 to May 2025; other languages, databases and grey literature may hold relevant work.
• Books, chapters and practice guidelines were excluded, and the authors single out book chapters as a possible source of foundational theory and extended case studies.
• Methodological immaturity in the included studies — conceptual work and small self-report samples — limits what can be concluded from the synthesis.
• The review does not claim conceptual saturation and describes its findings as provisional patterns.
• The article contains internal inconsistencies: a filtering section reports 3,340 records against 2,126 in the PRISMA flow, and the number of databases is given as six in the search-strategy section and seven elsewhere.

Connected Concepts

• Information and Media Literacy
• Libraries
• AI Literacy
• Librarians
• Curriculum Design
• Generative AI

Citation

Mirza, K. B., & Arif, M. (2026). Mapping the Integration of AI Literacy Into Information Literacy Instruction in Academic Libraries: A Systematic Review of Emerging Research. Education for Information, 1–37.